Legal claims defining the scope of protection, as filed with the USPTO.
1. A feature descriptor encoding apparatus comprising: a feature point detection unit configured to detect feature points from an image and outputs coordinate values; a local feature descriptor extracting unit configured to extract local feature descriptors from local regions centered on the coordinate values of the feature points; a coordinate value scanning unit configured to convert the coordinate values into index values according to a specific scanning method; a sorting unit configured to sort the index values; a difference calculation unit configured to calculate a differential value between two adjacent index values of the sorted index values and output a sequence of differential values; a differential encoding unit configured to encode the sequence of the differential values in a sequential permutation; and a local feature descriptor encoding unit configured to encode the local feature descriptors of the corresponding feature points in the same permutation as that of the sorted index values, wherein the differential encoding unit includes: a first encoding unit configured to perform M-bit fixed-length encoding by encoding the differential value when the differential value is in an encodable value area, and by encoding an escape code indicating that the differential value is outside the encodable value area when the differential value is outside the encodable value area; a second encoding unit configured to perform N-bit fixed length encoding by encoding the differential values by an encoding method different from that of the first encoding unit when the first encoding unit has encoded the escape code, N being larger than M; and an optimum encoding parameter determination unit configured to determine M and N from the sequence of the differential values such that a code length of the sequence of the encoded differential values is the shortest.
2. The feature descriptor encoding apparatus according to claim 1 , wherein the optimum encoding parameter determination unit is configured to calculate the code length while M and N are changed, and determine M and N such that the code length is the shortest.
3. The feature descriptor encoding apparatus according to claim 2 , wherein the optimum encoding parameter determination unit is configured to determine a minimum N based on an equation, N=ceil (log 2 (D MAX −2 M +2), where D MAX is a maximum value of the sequence of the differential values.
4. The feature descriptor encoding apparatus according to claim 3 , wherein the optimum encoding parameter determination unit is configured to determine M based on an equation, L=M×A+(M+N)×(Z−A), such that the code length is the shortest, where L, A, a function of detecting feature points from an image and outputting coordinate values; a function of extracting local feature descriptors from local regions centered on the coordinate values of the feature points; a function of converting the coordinate values into index values according to a specific scanning method; a function of sorting the index values; a function of calculating a differential value between two adjacent index values of the sorted index values and outputting a sequence of differential values; a function of encoding the sequence of the differential values in a sequential permutation; and a function of encoding the local feature descriptors of the corresponding feature points in the same permutation as that of the sorted index values, wherein said encoding of the sequence of the differential values includes: performing M-bit fixed-length encoding by encoding the differential value when the differential value is in an encodable value area, and by encoding an escape code indicating that the differential value is outside the encodable value area when the differential value is outside the encodable value area; performing N-bit fixed length encoding by encoding the differential values by an encoding method different from that of the first encoding unit when the first encoding unit has encoded the escape code, N being larger than M; and determining M and N from the sequence of the differential values such that a code length of the sequence of the encoded differential values is the shortest. and Z are the code length, a number of the differential values in the value area, and a number of the differential values in the sequence of the differential values, respectively.
5. A feature descriptor encoding method for causing a computer to execute: detecting feature points from an image and output coordinate values; extracting local feature descriptors from local regions centered on the coordinate values of the feature points; converting the coordinate values into index values according to a specific scanning method; sorting the index values; calculating a differential value between two adjacent index values of the sorted index values and output a sequence of differential values; encoding the sequence of the differential values in a sequential permutation; and encoding the local feature descriptors of the corresponding feature points in the same permutation as that of the sorted index values, wherein said encoding of the sequence of the differential values includes: performing M-bit fixed-length encoding by encoding the differential value when the differential value is in an encodable value area, and by encoding an escape code indicating that the differential value is outside the encodable value area when the differential value is outside the encodable value area; performing N-bit fixed length encoding by encoding the differential by encoding method different from that of the first encoding unit when the first encoding unit has encoded the escape code, N being larger than M; and determining M and N from the sequence of the differential values such that a code length of the sequence of the encoded differential values is the shortest.
6. A non-transitory computer-readable storage medium storing a program for causing a computer to realize.
Unknown
January 19, 2016
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.